Agent Sandbox Runner

Let coding agents actually boot the app and read the real error

Problem

When agents maintain many projects, they only read and write code — they can't boot the app, run migrations, hit the page, or read the actual error, so nobody knows if a change even works. Developers end up manually reproducing failures and feeding them back.

Solution

A self-hostable service that connects GitHub repos and spins up an isolated Docker/DDEV sandbox per project, runs the start script, migrations, and health checks, then feeds structured runtime logs and real errors back to the agent and the developer.

Users

Agencies and indie devs maintaining many client projects, small teams already using AI agents to batch-edit code

MVP

Connect GitHub repos and detect stack; generate per-repo sandbox from DDEV templates; one-click boot + migrate + health check; collect runtime logs and classify errors; tear down and replay sandboxes

Revenue

Open-source core plus hosted version billed by concurrent sandboxes or runtime minutes; private deployment license for agencies

Why now

Agents writing code is common, but the verification loop is still manual; containers and DDEV templates are mature enough for a solo dev to build the minimum loop fast